Since 2023, over 3,000 volunteers have searched satellite images for clouds; their answers are helping scientists check what computers miss

Satellites take thousands of pictures of Earth's atmosphere every day, making a huge collection of cloud movements over the globe. These images contain important information about weather patterns and climate processes, but identifying clouds accurately is not always as simple as it appears. Computers can sometimes have a problem identifying certain clouds and types of clouds, particularly when mixed with land masses, oceans, or other clouds and atmospheric effects. Since 2023, more than 3,000 volunteers have participated in the CloudCatcher project by looking at satellite images of clouds, assisting researchers in comparing their observations with the computer-aided approaches to cloud detection.

Why clouds are harder to read from space

Clouds can be seen as simple from the ground, but from space, they can be complex patterns of light, texture, and motion. Clouds vary in their composition, temperature, and impact on Earth's climate and weather according to their height. Clouds can reflect the sun out into space, or they can keep the heat in the atmosphere. Knowing where clouds are, and how they change and flow, is necessary to gain an understanding of the climate system of the Earth. But sometimes, satellite images aren't quite conclusive. The presence of shadows, highlights, thin cloud layers, and overlapping clouds can make it challenging for automated systems to accurately interpret what is being captured. The satellites collect massive amounts of information, and computer algorithms are necessary to process it and understand the information; accurate examples need to be built to test the performance of the systems. Here is where humans can come in handy.

The global effort behind better cloud detection

Researchers don’t have time to study all the satellite imagery that is being collected today. The CloudCatcher project did not solicit a small group of scientists to review all images, but rather it was an open call to the public. Volunteers were shown satellite images and asked to identify cloud types and locations. They provided observations that created a large body of human-labelled data, which could be compared with computer-generated results. The aim was not to dispense with the computer and turn to human helpers. Rather, observations of the human operation of the system gave scientists one more source of data that might enable them to better understand the strengths and weaknesses of automated systems.

How human judgment strengthened the artificial intelligence

Cloud identification is a major challenge in satellite research because computers learn from patterns in existing data. Automated systems can be influenced by training data that has errors and is not complete, repeating the errors in the analysis of new images. Volunteer classifications provided another way to examine how clouds appear in satellite images. These patterns could help researchers make predictions about cloud behaviour and improve cloud-detection methods. This is the way the CloudCatcher project aims to investigate if human observations can be used to improve satellite-based measurements. The volunteers’ classifications allow scientists to examine disagreements between people and computer systems and understand why those differences occur.

The growing power of citizen science

The project represents a growing use of citizen science in environmental research. Instead of limiting scientific analysis to professional laboratories, researchers can involve people around the world in collecting and organizing valuable information. It might appear that the task is just to view an image and identify the clouds. But thousands of individual observations together create a dataset that would require significant time and effort for a small research team to produce. The same method also allows people without scientific training to contribute directly to research that studies Earth’s changing environment. As satellites continue collecting larger amounts of atmospheric data, the challenge will not only be gathering images but also interpreting them accurately. These improved datasets can help current and future satellite systems make more accurate observations of Earth’s changing climate.